In this case study, we will use ProjecTILs to interpret human scRNA-seq T cell data in the context of a murine TIL atlas. We are going to use the single-cell dataset by Sade-Feldman GSE120575 to illustrate interspecies mapping of human cells on a stable murine atlas using gene orthologs.

Background

In the study by Sade-Feldman et al. (2018) Cell, the authours characterize the transcriptomics profile of immune cells from melanoma patients treated with checkpoint inhibitors, with the goal to identify factors that associate with success or failure of immune checkpoint therapy. They find that certain CD8+ T cell states associate with clinical outcome, and in particular that the transcription factor TCF7 is a significant marker for response to anti-PD-1 therapy. This suggests that the state of T cells found within a tumor is critical for induction of effective tumor immunity, highlighting the importance of characterizing (and potentially intervening on) the diversity of tumor-infiltraring T cells.

The single-cell expression data (GSE120575) consists of CD45+ single cells from 48 tumor samples of melanoma patients treated with checkpoint inhibitors, and sequenced by Smart-seq2 technology. Meta-data on response to therapy (Responder vs. Non-responder) is also available on the same GEO identifier.

Please use the development version of ProjecTILs >0.4.1, which implements ortholog gene conversion between human and mouse.

scRNA-seq data preparation

Download the count matrix and metadata from Gene Expression Omnibus (GEO), and store as Seurat object.

Some basic statistics - cells per group (Pre vs. Post, Therapy, Responder vs. Non-responder)


 Post   Pre 
10363  5928 

    anti-CTLA4 anti-CTLA4+PD1       anti-PD1 
           517           4121          11653 

Non-responder     Responder 
        10727          5564 

ProjecTILs

Load reference TIL atlas - if it’s not present in the working directory, it will be downloaded from the repository

[1] "Loading Default Reference Atlas..."
[1] "/Users/mass/Documents/Projects/Github/ProjecTILs_CaseStudies/ref_TILAtlas_mouse_v1.rds"
[1] "Loaded Reference map ref_TILAtlas_mouse_v1"

Run projection algorithm - version 0.4.1 supports human ortholog conversion (human.ortho=T)

[1] "Using assay RNA for query"
[1] "4862 out of 16291 ( 30 % ) non-pure T cells removed.  Use filter.cells=FALSE to avoid pre-filtering (NOT RECOMMENDED)"
[1] "Transforming expression matrix into space of mouse orthologs"
[1] "Aligning query to reference map for batch-correction..."

Projecting corrected query onto Reference PCA space
[1] "Projecting corrected query onto Reference UMAP space"

Plot global projection of new data over the reference in UMAP space

Predict the cell states in the query set


     CD4_NaiveLike     CD8_EarlyActiv CD8_EffectorMemory      CD8_NaiveLike 
               187               1181               2153               2119 
           CD8_Tex           CD8_Tpex                Tfh                Th1 
              1594                675                922               1523 
              Treg 
              1075 

The expression profiles or some key genes correspond fairly well to those of the reference

Note that projections and comparisons are performed in the ortholog space of murine genes - to check the names of human-mouse orthologs you can examine the conversion table for genes of interest:

      Gene.stable.ID.HS Gene.HS Gene.MM                         Alt.symbol
7381    ENSG00000254126    CD8B   Cd8b1                              Cd8b1
7835    ENSG00000188389   PDCD1   Pdcd1                              Pdcd1
11096   ENSG00000081059    TCF7    Tcf7                               Tcf7
16688   ENSG00000100453    GZMB    Gzmb Gzmd,Gzme,Gzmf,Gzmc,Gzmn,Gzmg,Gzmb
      Alt.symbol.HS
7381     CD8B2,CD8B
7835          PDCD1
11096          TCF7
16688          GZMB

Response to therapy

Now to the interesting part. Let’s visualize the projection and T cell state distribution by response to therapy:

There is a clear shift to more differentiated states (e.g. CD8_Tex and CD8_EM) in non-responders, while responders have a higher fraction of Naive-like states.

To better examine the different in T cell phenotype composition between rsponders and non-responders, we can visualize the fold-change of T cell state frequency between the two groups:

CD4 and CD8 Naive-like states are the most enriched in responders compared to non-responders, confirming the observation of the original paper that TCF7+ cells are associated with response to therapy. Additionally, we also observe an increase in CD4+ T follicular helper cells (Tfh), as well as a decrease in effector CD8+ cells. As expected, T regulatory cells (Tregs) are more prevalent in non-responders. Because Tregs are known to inhibit antitumor immunuty, this observation that could also explain the reduced tumor control in non-responding lesions.

Conclusions

Taking advantage of the ortholog mapping functionality of ProjecTILs, we have illustrated how to effortlessy analyze human scRNA-seq data in the context of a reference murine TIL atlas. Gene expression profiles confirm that T cells are accurately projected in major CD4+ and CD8+ categories, as well as in more specific subtypes. Comparing the transcriptomics profile of immune cells from melanoma patients treated with checkpoint inhibitors, and their composition in terms of cell states, confirms the observation of the original study that TCF7 is a marker and potential predictor of immunotherapy response. Additionally, ProjecTILs analysis in the context of a stable atlas gives a more complete picture of the T cell states that are “favorable” to tumor response - e.g. the enrichment of more naive-like states, and reduction of Tregs within the tumor.

Further reading

Original publication - Sade-Feldman et al. (2018) Cell

ProjecTILs case studies - INDEX - Repository